wrong result when using zero copy input

Hello Xu, 

Please provide the part number of the processor you are using. 

Thank you.

~Leonard 

TDA4VH-q1.
Thank you.

Hi,

Which sdk version you are using ?

8.6

To cross out the SDK version compatibility issue, can you share observations on 9.1 SDK ? 

The recent sdk has multiple fixes on model inference side, which potentially solve this issue.

Thank you for your reply. I will test it on version 9.1

By the way, does it cause the issue if the first layer of the model is a convolution layer, but the second layer is a resize layer?

Noted,

Can you comment on the layer structures for models where enable padding param is working fine and model where its not working.

First layer dimensions, followed by next 2 to 3 layer info would be good.

Thank you 

The layer structures in the picture above is the model which is not working. The first convolution layer was added to use the configuration "inDataPadInTidl=1". However it dose not work.

The structure in this picture is one of the normal models. 

Thank you~

Thanks for the info.

Look forward to 9.1 sdk results